Professional Documents
Culture Documents
Philosophy And Model Theory Tim Button full chapter pdf docx
Philosophy And Model Theory Tim Button full chapter pdf docx
Button
Visit to download the full and correct content document:
https://ebookmass.com/product/philosophy-and-model-theory-tim-button/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
https://ebookmass.com/product/model-predictive-control-theory-
computation-and-design-2nd-edition-rawlings-james-b/
https://ebookmass.com/product/philosophy-of-film-without-theory-
craig-fox/
https://ebookmass.com/product/gender-issues-and-philosophy-
education-history-theory-practice-markus-tiedemann/
https://ebookmass.com/product/the-art-of-anatheism-reframing-
continental-philosophy-of-religion-1st-edition-tim-di-muzio/
The Philosophy of Evolutionary Theory: Concepts,
Inferences, and Probabilities 1st Edition Elliott Sober
https://ebookmass.com/product/the-philosophy-of-evolutionary-
theory-concepts-inferences-and-probabilities-1st-edition-elliott-
sober/
https://ebookmass.com/product/the-politics-of-vulnerable-groups-
implications-for-philosophy-law-and-political-theory-fabio-
macioce/
https://ebookmass.com/product/indonesian-law-tim-lindsey/
https://ebookmass.com/product/boxing-the-octopus-tim-maleeny/
https://ebookmass.com/product/landscape-materiality-and-heritage-
an-object-biography-tim-edensor/
Philosophy and Model Theory
Philosophy and
Model Theory
Philosophy and model theory frequently meet one another. This book aims to un-
derstand their interactions.
Model theory is used in every ‘theoretical’ branch of analytic philosophy: in phi-
losophy of mathematics; in philosophy of science; in philosophy of language; in
philosophical logic; and in metaphysics. But these wide-ranging appeals to model
theory have created a highly fragmented literature. On the one hand, many philo-
sophically significant results are found only in mathematics textbooks: these are
aimed squarely at mathematicians; they typically presuppose that the reader has a
serious background in mathematics; and little clue is given as to their philosophical
significance. On the other hand, the philosophical applications of these results are
scattered across disconnected pockets of papers.
The first aim of our book, then, is to consider the philosophical uses of model the-
ory. We state and prove the best versions of results for philosophical purposes. We
then probe their philosophical significance. And we show how similar dialectical
situations arise repeatedly across fragmented debates in different areas.
The second aim of our book, though, is to consider the philosophy of model theory.
Model theory itself is rarely taken as the subject matter of philosophising (contrast
this with the philosophy of biology, or the philosophy of set theory). But model
theory is a beautiful part of pure mathematics, and worthy of philosophical study
in its own right.
Both aims give rise to challenges. On the one hand: the philosophical uses of
model theory are scattered across a disunified literature. And on the other hand:
there is scarcely any literature on the philosophy of model theory.
All of which is to say: philosophy and model theory isn’t really ‘a thing’ yet. This
book aims to start carving out such a thing. We want to chart the rock-face and
trace its dialectical contours. But we present this book, not as a final word on what
philosophically inclined model theorists and model-theoretically inclined philoso-
phers should do, but as an invitation to join in.
So. This is not a book in which a single axe is ground, page by page, into an
increasingly sharp blade. No fundamental line of argument—arching from Chapter
1 through to Chapter 17—serves as the spine of the book. What knits the chapters
together into a single book is not a single thesis, but a sequence of overlapping,
criss-crossing themes.
vi preface
Topic selection
Precisely because philosophy and model theory isn’t yet ‘a thing’, we have had to
make some difficult decisions about what topics to discuss.
On the one hand, we aimed to pick topics which should be of fairly mainstream
philosophical concern. So, when it comes to the philosophical uses of model the-
ory, we have largely considered topics concerning reference, realism, and doxology
(a term we introduce in Chapter 6). But, even when we have considered questions
which fall squarely within in the philosophy of model theory, the questions that
we have focussed on are clearly instances of ‘big questions’. We look at questions
of sameness of theories/structure (Chapter 5); of taking diverse perspectives on the
same concept (in Chapter 14); of how to draw boundaries of logic (in Chapter 16);
and of classification of mathematical objects (Chapter 17).
On the other hand, we also wanted to give you a decent bang for your buck. We
figured that if you were going to have to wrestle with some new philosophical idea
or model-theoretic result, then you should get to see it put to decent use. (This ex-
plains why, for example, the Push-Through Construction, the just-more theory ma-
noeuvre, supervaluational semantics, and the ideas of moderation and modelism,
occur so often in this book.) Conversely, we have had to set aside debates—no
matter how interesting—which would have taken too long to set up.
All of which is to say: this book is not comprehensive. Not even close.
Although we consider models of set theory in Chapters 8 and 11, we scarcely
scratch the surface. Although we discuss infinitary logics in Chapters 15–16, we only
use them in fairly limited ways.1 Whilst we mention Tennenbaum’s Theorem in
Chapter 7, that is as close as we get to computable model theory. We devote only
one brief section to o-minimality, namely §4.10. And although we consider quanti-
fiers in Chapter 16 and frequently touch on issues concerning logical consequence,
we never address the latter topic head on.2
The grave enormity, though, is that we have barely scratched the surface of model
theory itself. As the table of contents reveals, the vast majority of the book considers
model theory as it existed before Morley’s Categoricity Theorem.
Partially correcting for this, Wilfrid Hodges’ wonderful historical essay appears
as Part D of this book. Wilfrid’s essay treats Morley’s Theorem as a pivot-point for
the subject of model theory. He looks back critically to the history, to uncover the
notions at work in Morley’s Theorem and its proof, and he looks forward to the
riches that have followed from it. We have both learned so much from Wilfrid’s
Model Theory,3 and we are delighted to include his ‘short history’ here.
1 We do not, for example, consider the connection between infinitary logics and supervenience, as
Still, concerning all those topics which we have omitted: we intend no slight
against them. But there is only so much one book can do, and this book is already
much (much) longer than we originally planned. We are sincere in our earlier claim,
that this book is not offered as a final word, but as an invitation to take part.
day); proofs which are difficult to access in the existing literature; proofs of certain
folk-lore results; and proofs of new results.
But the book omits all proofs which are both readily accessed and too long to be
self-contained. In such cases, we simply provide readers with citations.
The quick moral for readers to extract is this. If you encounter a proof in the main
text of a chapter, you should follow it through. But we would add a note for read-
ers whose primary background is in philosophy. If you really want to understand a
mathematical concept, you need to see it in action. Read the appendices!
Acknowledgements
The book arose from a seminar series on philosophy and model theory that we ran
in Birkbeck in Autumn 2011. We turned the seminar into a paper, but it was vastly
too long. An anonymous referee for Philosophia Mathematica suggested the paper
might form the basis for a book. So it did.
We have presented topics from this book several times. It would not be the
book it is, without the feedback, questions and comments we have received. So
we owe thanks to: an anonymous referee for Philosophia Mathematica, and James
Studd for OUP; and to Sarah Acton, George Anegg, Andrew Arana, Bahram Assa-
dian, John Baldwin, Kyle Banick, Neil Barton, Timothy Bays, Anna Bellomo, Liam
Bright, Chloé de Canson, Adam Caulton, Catrin Campbell-Moore, John Corco-
ran, Radin Dardashti, Walter Dean, Natalja Deng, William Demopoulos, Michael
Detlefsen, Fiona Doherty, Cian Dorr, Stephen Duxbury, Sean Ebels-Duggan, Sam
Eklund, Hartry Field, Branden Fitelson, Vera Flocke, Salvatore Florio, Peter Fritz,
Michael Gabbay, Haim Gaifman, J. Ethan Galebach, Marcus Giaquinto, Peter Gib-
son, Tamara von Glehn, Owen Griffiths, Emmylou Haffner, Bob Hale, Jeremy
Heis, Will Hendy, Simon Hewitt, Kate Hodesdon, Wilfrid Hodges, Luca Incur-
vati, Douglas Jesseph, Nicholas Jones, Peter Koellner, Brian King, Eleanor Knox, Jo-
hannes Korbmacher, Arnold Koslow, Hans-Christoph Kotzsch, Greg Lauro, Sarah
Lawsky, Øystein Linnebo, Yang Liu, Pen Maddy, Kate Manion, Tony Martin, Guil-
laume Massas, Vann McGee, Toby Meadows, Richard Mendelsohn, Christopher
Mitsch, Stella Moon, Adrian Moore, J. Brian Pitts, Jonathan Nassim, Fredrik Ny-
seth, Sara Parhizgari, Charles Parsons, Jonathan Payne, Graham Priest, Michael
Potter, Hilary Putnam, Paula Quinon, David Rabouin, Erich Reck, Sam Roberts,
Marcus Rossberg, J. Schatz, Gil Sagi, Bernhard Salow, Chris Scambler, Thomas
Schindler, Dana Scott, Stewart Shapiro, Gila Sher, Lukas Skiba, Jönne Speck, Se-
bastian Speitel, Will Stafford, Trevor Teitel, Robert Trueman, Jouko Väänänen, Kai
Wehmeier, J. Robert G. Williams, John Wigglesworth, Hugh Woodin, Jack Woods,
Crispin Wright, Wesley Wrigley, and Kino Zhao.
We owe some special debts to people involved in the original Birkbeck seminar.
preface ix
First, the seminar was held under the auspices of the Department of Philosophy at
Birkbeck and Øystein Linnebo’s European Research Council-funded project ‘Plu-
rals, Predicates, and Paradox’, and we are very grateful to all the people from the
project and the department for participating and helping to make the seminar pos-
sible. Second, we were lucky to have several great external speakers visit the sem-
inar, whom we would especially like to thank. The speakers were: Timothy Bays,
Walter Dean, Volker Halbach, Leon Horsten, Richard Kaye, Jeff Ketland, Angus
Macintyre, Paula Quinon, Peter Smith, and J. Robert G. Williams. Third, many of
the external talks were hosted by the Institute of Philosophy, and we wish to thank
Barry C. Smith and Shahrar Ali for all their support and help in this connection.
A more distant yet important debt is owed to Denis Bonnay, Brice Halimi, and
Jean-Michel Salanskis, who organised a lovely event in Paris in June 2010 called ‘Phi-
losophy and Model Theory.’ That event got some of us first thinking about ‘Philos-
ophy and Model Theory’ as a unified topic.
We are also grateful to various editors and publishers for allowing us to reuse pre-
viously published material. Chapter 5 draws heavily on Walsh 2014, and the copy-
right is held by the Association for Symbolic Logic and is being used with their
permission. Chapters 7–11 draw heavily upon on Button and Walsh 2016, published
by Philosophia Mathematica. Finally, §13.7 draws from Button 2016b, published by
Analysis, and §15.1 draws from Button 2017, published by the Notre Dame Journal of
Formal Logic.
Finally, though, a word from us, as individuals.
From Tim. I want to offer my deep thanks to the Leverhulme Trust: their fund-
ing, in the form of a Philip Leverhulme Prize (plp–2014–140), enabled me to take
the research leave necessary for this book. But I mostly want to thank two very spe-
cial people. Without Sean, this book could not be. And without my Ben, I could
not be.
From Sean. I want to thank the Kurt Gödel Society, whose funding, in the form
of a Kurt Gödel Research Prize Fellowship, helped us put on the original Birkbeck
seminar. I also want to thank Tim for being a model co-author and a model friend.
Finally, I want to thank Kari for her complete love and support.
Contents
B Categoricity 137
15 Indiscernibility 359
15.1 Notions of indiscernibility . . . . . . . . . . . . . . . . . . . . . . 359
15.2 Singling out indiscernibles . . . . . . . . . . . . . . . . . . . . . . 366
15.3 The identity of indiscernibles . . . . . . . . . . . . . . . . . . . . . 370
15.4 Two-indiscernibles in infinitary logics . . . . . . . . . . . . . . . . 376
15.5 n-indiscernibles, order, and stability . . . . . . . . . . . . . . . . . 380
15.a Charting the grades of discernibility . . . . . . . . . . . . . . . . . 384
16 Quantifiers 387
16.1 Generalised quantifiers . . . . . . . . . . . . . . . . . . . . . . . . 387
16.2 Clarifying the question of logicality . . . . . . . . . . . . . . . . . 389
16.3 Tarski and Sher . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
16.4 Tarski and Klein’s Erlangen Programme . . . . . . . . . . . . . . . 390
16.5 The Principle of Non-Discrimination . . . . . . . . . . . . . . . . 392
16.6 The Principle of Closure . . . . . . . . . . . . . . . . . . . . . . . 399
16.7 McGee’s squeezing argument . . . . . . . . . . . . . . . . . . . . . 407
16.8 Mathematical content . . . . . . . . . . . . . . . . . . . . . . . . . 408
16.9 Explications and pluralism . . . . . . . . . . . . . . . . . . . . . . 410
18 Wilfrid Hodges
A short history of model theory 439
18.1 ‘A new branch of metamathematics’ . . . . . . . . . . . . . . . . . 439
18.2 Replacing the old metamathematics . . . . . . . . . . . . . . . . . 440
18.3 Definable relations in one structure . . . . . . . . . . . . . . . . . 445
18.4 Building a structure . . . . . . . . . . . . . . . . . . . . . . . . . . 449
18.5 Maps between structures . . . . . . . . . . . . . . . . . . . . . . . 455
18.6 Equivalence and preservation . . . . . . . . . . . . . . . . . . . . . 460
18.7 Categoricity and classification theory . . . . . . . . . . . . . . . . 465
18.8 Geometric model theory . . . . . . . . . . . . . . . . . . . . . . . 469
18.9 Other languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472
18.10 Model theory within mathematics . . . . . . . . . . . . . . . . . . 474
18.11 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475
18.12 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . 475
Bibliography 477
Index 507
they should know that this section assumes a prior understanding of §§2.1, 4.1, 4.2,
and much (but not all) of Chapter 1. (We omit purely technical appendices from
this diagram.)
1.1
1.2
1.3
1.5 1.4
1.7 1.6
1.9 1.8
1.10
2.1 1.11
2.2 1.12 1.a
2.3 3.1 1.13 1.b 1.c
2.4 2.a 3.2 4.1
2.5 3.3 4.2
3.5 4.3
3.6 4.4
5.1 3.4 3.7 4.5 4.7
5.2 5.3 3.8 4.6
5.4 3.9 4.8
5.5 3.10 3.a 4.9
5.6 5.7 5.8 4.10
1
Logics and languages
Model theory begins by considering the relationship between languages and struc-
tures. This chapter outlines the most basic aspects of that relationship.
One purpose of the chapter will therefore be immediately clear: we want to lay
down some fairly dry, technical preliminaries. Readers with some familiarity with
mathematical logic should feel free to skim through these technicalities, as there are
no great surprises in store.
Before the skimming commences, though, we should flag a second purpose of
this chapter. There are at least three rather different approaches to the semantics
for formal languages. In a straightforward sense, these approaches are technically
equivalent. Most books simply choose one of them without comment. We, how-
ever, lay down all three approaches and discuss their comparative strengths and
weaknesses. Doing this highlights that there are philosophical discussions to be had
from the get-go. Moreover, by considering what is invariant between the different
approaches, we can better distinguish between the merely idiosyncratic features of
a particular approach, and the things which really matter.
One last point, before we get going: tradition demands that we issue a caveat.
Since Tarski and Quine, philosophers have been careful to emphasise the impor-
tant distinction between using and mentioning words. In philosophical texts, that
distinction is typically flagged with various kinds of quotation marks. But within
model theory, context almost always disambiguates between use and mention.
Moreover, including too much punctuation makes for ugly text. With this in mind,
we follow model-theoretic practice and avoid using quotation marks except when
they will be especially helpful.
Definition 1.1: A signature, L , is a set of symbols, of three basic kinds: constant sym-
bols, relation symbols, and function symbols. Each relation symbol and function symbol
has an associated number of places (a natural number), so that one may speak of an
n-place relation or function symbol.
8 logics and languages
Throughout this book, we use script fonts for signatures. Constant symbols should
be thought of as names for entities, and we tend to use c 1 , c 2 , etc. Relation symbols,
which are also known as predicates, should be thought of as picking out properties
or relations. A two-place relation, such as x is smaller than y, must be associated
with a two-place relation symbol. We tend to use R1 , R2 , etc. for relation symbols.
Function symbols should be thought of as picking out functions and, again, they
need an associated number of places: the function of multiplication on the natural
numbers takes two natural numbers as inputs and outputs a single natural number,
so we must associate that function with a two-place function symbol. We tend to
use f 1 , f 2 , etc. for function symbols.
The examples just given—being smaller than, and multiplication on the natural
numbers—suggest that we will use our formal vocabulary to make determinate
claims about certain objects, such as people or numbers. To make this precise,
we introduce the notion of an L -structure; that is, a structure whose signature is
L . An L -structure, M, is an underlying domain, M, together with an assignment
of L ’s constant symbols to elements of M, of L ’s relation symbols to relations
on M, and of L ’s function symbols to functions over M. We always use calligraphic
fonts M, N, … for structures, and M, N, … for their underlying domains. Where s
is any L -symbol, we say that s M is the object, relation or function (as appropriate)
assigned to s in the structure M. This informal explanation of an L -structure is
always given a set-theoretic implementation, leading to the following definition:
a ∈ A and b ∈ B}. Likewise, we recursively define ordered n-tuples in terms of ordered pairs by setting e.g.
(a, b, c) = ((a, b), c).
1.2. first-order logic: a first look 9
Definition 1.3: The following, and nothing else, are first-order L -terms:
• any variable, and any constant symbol c from L
2 A pedantic comment is in order. The symbols ‘t ’ and ‘t ’ are not being used here as expressions in
1 2
the object language (i.e. first-order logic with signature L ). Rather, they are being used as expressions of
the metalanguage, within which we describe the syntax of first-order L -terms and L -formulas. Similarly,
the symbol ‘x’, as it occurs in the last clause of Definition 1.3, is not being used as an expression of the object
language, but in the metalanguage. So the final clause in this definition should be read as saying something
like this. For any variable and any formula φ which does not already contain a concatenation of a quantifier
followed by that variable, the following concatenation is a formula: a quantifier, followed by that variable,
followed by φ. (The reason for this clause is to guarantee that e.g. ∃v∀vF(v) is not a formula.) We could
flag this more explicitly, by using a different font for metalinguistic variables (for example). However, as
with flagging quotation, we think the additional precision is not worth the ugliness.
10 logics and languages
fact, there are many different but extensionally equivalent approaches to defining
this relation, and we will consider three in this chapter.
To understand why there are several different approaches to the semantics for
first-order logic, we must see why the most obvious approach fails. Our sentences
have a nice, recursive syntax, so we will want to provide them with a nice, recursive
semantics. The most obvious starting point is to supply semantic clauses for the
two kinds of atomic sentence, as follows:
M ⊧ t 1 = t 2 iff t M
1 = t2
M
M ⊧ R(t 1 , …, t n ) iff (t M
1 , …, t n ) ∈ R
M M
Next, we would need recursion clauses for the quantifier-free sentences. So, writing
M ⊭ φ for it is not the case that M ⊧ φ, we would offer:
M ⊧ ¬φ iff M ⊭ φ
M ⊧ (φ ∧ ψ) iff M ⊧ φ and M ⊧ ψ
So far, so good. But the problem arises with the quantifiers. Where the notation
φ(c/x) indicates the formula obtained by replacing every instance of the free vari-
able x in φ(x) with the constant symbol c, an obvious thought would be to try:
Unfortunately, this recursion clause is inadequate. To see why, suppose we had a very
simple signature containing a single one-place predicate R and no constant symbols.
Then, for any structure M in that signature, we would vacuously have that M ⊧
∀vR(v). But this would be the case even if RM = ∅, that is, even if nothing had the
property picked out by R. Intuitively, that is the wrong verdict.
The essential difficulty in defining the semantics for first-order logic therefore
arises when we confront quantifiers. The three approaches to semantics which we
consider present three ways to overcome this difficulty.
investment gives a decent philosophical pay-off. For, as we move through the chap-
ter, we will see that these (quite dry) technicalities can both generate and resolve
philosophical controversies.
Third: we expect that even novice philosophers reading this book will have at
least a rough and ready idea of what is coming next. And such readers will be bet-
ter served by reading (and perhaps only partially absorbing) multiple different ap-
proaches to the semantics for first-order logic, than by trying to rote-learn one spe-
cific definition. They will thereby get a sense of what is important to supplying a
semantics, and what is merely an idiosyncratic feature of a particular approach.
To illustrate this definition, suppose that M is the natural numbers in the signature
{0, 1, +, ×}, with each symbol interpreted as normal. (This licenses us in dropping
the ‘M’-superscript when writing the symbols.) Suppose that σ and τ are variable-
assignments such that σ(x1 ) = 5, σ(x2 ) = 7, τ(x1 ) = 3, τ(x2 ) = 7, and consider
the term t(x1 , x2 ) = (1 + x1 ) × (x1 + x2 ). Then we can compute the interpretation
of the term relative to the variable-assignments as follows:
t M,σ = (1 + xM,σ
1 ) × (x1
M,σ
+ xM,σ
2 ) = (1 + 5) × (5 + 7) = 72
t M,τ = (1 + xM,τ
1 ) × (x1
M,τ
+ xM,τ
2 ) = (1 + 3) × (3 + 7) = 40
3 See Tarski (1933) and Tarski and Vaught (1958), but also §12.a.
1.4. semantics for variables 13
M, σ ⊧ t 1 = t 2 iff t M,σ
1 = t M,σ
2 , for any L -terms t 1 , t 2
We leave it to the reader to formulate clauses for disjunction and existential quan-
tification. Finally, where φ is any first-order L -sentence, we say that M ⊧ φ iff
M, σ ⊧ φ for all variable-assignments σ.
4 Quine (1981: §12). For ease of reference, we cite the 1981-edition. However, the relevant sections
are entirely unchanged from the (first) 1940-edition. We owe several people thanks for discussion of ma-
terial in this section. Michael Potter alerted us to Bourbaki’s notation; Kai Wehmeier alerted us to Quine’s
(cf. Wehmeier forthcoming); and Robert Trueman suggested that we should connect all of this to Fine’s
antinomy of the variable.
5 Fine (2003: 606, 2007: 7), for this and all subsequent quotes from Fine.
14 logics and languages
But, as Fine notes, this cannot be right either. For, ‘when we consider the semantic
role of the variables in the same expression—such as “x > y”—then it seems equally
clear that their semantic role is different.’ So Fine says:
(b) ‘Any two variables (ranging over a given domain of objects) have a different
semantic role.’
And now we have arrived at Fine’s antinomy of the variable.
We think that this whole antinomy gets going from the mistaken assumption that
we can assign a ‘semantic role’ to a variable in isolation from the quantifier which
binds it.6 As Quine put the point more than six decades before Fine: ‘The vari-
ables […] serve merely to indicate cross-references to various positions of quantifi-
cation.’7 Quine’s point is that ∃x∀yφ(x, y) and ∃y∀xφ(y, x) are indeed just typo-
graphical variants, but that both are importantly different from ∀x∃yφ(x, y). And
to illustrate this graphically, Quine notes that we could use a notation which aban-
dons typographically distinct variables altogether. For example, instead of writing:
Bourbaki rigorously developed Quine’s brief notational suggestion.9 And the re-
sulting Quine–Bourbaki notation is evidently just as expressively powerful as our or-
dinary notation. However, if we adopt the Quine–Bourbaki notation, then we will
not even be able to ask whether typographically distinct variables like ‘x’ and ‘y’ have
different ‘semantic roles’, and Fine’s antinomy will dissolve away.10
6 Fine (2003: 610–14, 2007: 12–16) considers this thought, but does not consider the present point.
7 Quine (1981: 69–70). See also Curry (1933: 389–90), Quine (1981: iv, 5, 71), Dummett (1981: ch.1),
Kaplan (1986: 244), Lavine (2000: 5–6), and Potter (2000: 64).
8 Quine (1981: §12).
9 Bourbaki (1954: ch.1), apparently independently. The slight difference is that Bourbaki uses
nomy. Pickel and Rabern assume that the Quine–Bourbaki approach will be coupled with Frege’s idea that
one obtains the predicate ‘( ) ≤ ( )’ by taking a sentence like ‘7 ≤ 7’ and deleting the names. They then
insist that Frege must distinguish between the case when ‘( ) ≤ ( )’ is regarded as a one-place predicate,
and the case where it is regarded as a two-place predicate. And they then maintain: ‘if Frege were to intro-
duce marks capable of typographically distinguishing between these predicates, then that mark would need
its own semantic significance, which in this context means designation.’ We disagree with the last part of
this claim. Brackets are semantically significant, in that ¬(φ ∧ ψ) is importantly different from (¬φ ∧ ψ);
but brackets do not denote. Fregeans should simply insist that any ‘marks’ on predicate-positions have a
similarly non-denotational semantic significance. After all, their ultimate purpose is just to account for the
different ‘cross-referencing’ in ∀x∃yφ(x, y) and ∀x∃yφ(y, x).
1.5. the robinsonian approach to semantics 15
In Quinean terms, the difference between a model and its reduct is not ontological
but ideological.14 We do not add or remove any entities from the domain; we just
add or remove some (interpretations of) symbols.
11 Where φ is any Quine–Bourbaki sentence, let φ fo be the sentence of first-order logic which results
by: (a) inserting the variable vn after the nth quantifier in φ, counting quantifiers from left-to-right; (b)
replacing each blob connected to the nth -quantifier with the variable vn and (c) deleting all the connecting
wires. Then say M ⊧ φ iff M ⊧ φfo , with M ⊧ φfo defined via the Tarskian approach.
12 A. Robinson (1951: 19–21), with a tweak that one finds in, e.g., Sacks (1972: ch.4).
13 Cf. Hodges (1993: 9ff) and Marker (2002: 31).
14 Quine (1951: 14).
16 logics and languages
We can now define the idea of ‘adding new constants for every member of the
domain’. The following definition explains how to add, for each element a ∈ M, a
new constant symbol, c a , which is taken to name a:
Definition 1.5: Let L be any signature. For any set M, L (M) is the signature obtained
by adding to L a new constant symbol c a for each a ∈ M. For any L -structure M with
domain M, we say that M○ is the L (M)-structure whose L -reduct is M and such that
○
a = a for all a ∈ M.
cM
Since M○ is flooded with constants, it is very easy to set up its semantics. We start
by defining the interpretation of the L (M)-terms which contain no variables:
○ ○ ○ ○
1 , …, s k ), if t is the variable-free L (M)-term f (s 1 , …, s k )
t M = f M (s M M
○ ○ ○
M○ ⊧ R(t 1 , …, t n ) iff (t M
1 , …, t n ) ∈ R
M M
, for
any variable-free L (M)-terms t 1 , …, t n and
any n-place relation symbol R from L (M)
M○ ⊧ ¬φ iff M○ ⊭ φ
M○ ⊧ (φ ∧ ψ) iff M○ ⊧ φ and M○ ⊧ ψ
M○ ⊧ ∀xφ(x) iff M○ ⊧ φ(c a /x) for every a ∈ M
This completes the Robinsonian approach. And the approach carries no taint of
the antinomy of the variable, since it clearly accords variables with no more seman-
tic significance than is suggested by the Quine–Bourbaki notation. Indeed, it is
easy to give a Robinsonian semantics directly for Quine–Bourbaki sentences, via:
M○ satisfies a Quine–Bourbaki sentence beginning with ‘∀’ iff for every a ∈ M the
model M○ satisfies the Quine–Bourbaki sentence which results from replacing all
blobs connected to the quantifier with ‘c a ’ and then deleting the quantifier and the
connecting wires.
true in W using Robinson’s own proposal, we would have to treat each wombat as a
name for itself, and so imagine a language whose syntactic parts are live wombats.17
This stretches the ordinary notion of a language to breaking point.
There is also a technical hitch with Robinson’s own proposal. Suppose that c is
a constant symbol of L . Suppose that M is an L -structure where the symbol c
is itself an element of M’s underlying domain. Finally, suppose that M interprets c
as naming some element other than c itself, i.e. c M ≠ c. Now Robinson’s proposal
○
requires that c M = c. But since M○ is a signature expansion of M, we require that
○
c M = c M , which is a contradiction.
To fix this bug whilst retaining Robinson’s idea that c a = a, we would have to
tweak the definition of an L -structure to ensure that the envisaged situation can-
not arise.18 A better alternative—which also spares the wombats—is to abandon
Robinson’s suggestion that c a = a, and instead define the symbol c a so that it is
guaranteed not to be an element of M’s underlying domain.19 So this is our official
Robinsonian semantics (even if it was not exactly Robinson’s).
M ⊧ R(t 1 , …, t n ) iff (t M
1 , …, t n ) ∈ R , for any variable-free L -terms
M M
All that remains is to define M ⊧ φ(a) without going all-out Robinsonian. And the
idea here is quite simple: we just add new constant symbols when we need them,
but not before. Here is the idea, rigorously developed. Let M be an L -structure
with a from M. For each ai among a, let c ai be a constant symbol not occurring in
L . Intuitively, we interpret each c ai as a name for ai . More formally, we define
M[a] to be a structure whose signature is L together with the new constant sym-
M[a]
bols among c a , whose L -reduct is M, and such that c ai = ai for each i. Where
φ(v) is an L -formula with free variables displayed, the Hybrid approach defines:
When we combine our new definition of M ⊧ φ(a) with the clause for universal
quantification, we see that universal quantification effectively amounts to consider-
ing all the different ways of expanding the signature of M with a new constant sym-
bol which could be interpreted to name any element of M. (So the Hybrid approach
offers a semantics by simultaneous recursion over structures and languages.) Fi-
nally, we offer a similar clause for terms:
similar. But the clearest examples we can find are Boolos and Jeffrey (1974: 104–5), Boolos (1975: 513–4),
and Lavine (2000: 10–12).
21 See Lavine’s (2000: 12–13) comments on compositionality and learnability.
20 logics and languages
if a language has this feature, then we ‘understand how an infinite aptitude can be
encompassed by finite accomplishments’. Conversely, if ‘a language lacks this fea-
ture then no matter how many sentences a would-be speaker learns to produce and
understand, there will remain others whose meanings are not given by the rules
already mastered.’22
Davidson’s argument is too quick. After all, it is a wild idealisation to suggest that
any actual human can indeed understand or learn the meanings of infinitely many
sentences: some sentences are just too long for any actual human to parse. It is
unclear, then, why we should worry about the ‘learnability’ of such sentences.
Still, something in the vicinity of Davidson’s argument seems right. In §1.6, we
floated the idea that model theory should be regarded as a branch of applied math-
ematics, whose (idealised) subject matter is the languages and theories that (pure)
mathematicians actually use. But here is an apparent phenomenon concerning that
subject matter: once we have a fixed interpretation in mind, we tend to act as if that
interpretation fixes the truth value of any sentence of the appropriate language, no
matter how long or complicated that sentence is.23 All three of our approaches to
formal semantics accommodate this point. For, given a signature L and an L -
structure M—i.e. an interpretation of the range of quantification and an interpre-
tation of each L -symbol—the semantic value of every L -sentence is completely
determined within M, in the sense that, for every L -sentence φ, either M ⊧ φ, or
M ⊧ ¬φ, but not both.
But the Hybrid approach, specifically, may allow us to go a little further. For,
when L is finite,24 and we offer the Hybrid approach to semantics, we may gain
some insight into how a finite mind might fully understand the rules by which an
interpretation fixes the truth-value of every sentence. That understanding seems to
reduce to three rather tractable components:
(a) an understanding of the finitely many recursion clauses governing satisfac-
tion for atomic sentences (finitely many, as we assumed that L is finite);
(b) an understanding of the handful of recursion clauses governing sentential
connectives; and
(c) an understanding of the recursion clauses governing quantification
On the Hybrid approach, point (c) reduces to an understanding of two ideas: (i)
the general idea that names can pick out objects,25 and (ii) the intuitive idea that, for
any object, we could expand our language with a new name for that object. In short:
Chapters 2 and 15). However, the general notion seems to be required by any model-theoretic semantics,
so that there is no special problem here for the Hybrid approach.
1.9. second-order logic: syntax 21
the Hybrid semantics seems to provide a truly compositional notion of meaning. But
we should be clear on what this means.
First, we are not aiming to escape what Sheffer once called the ‘logocentric
predicament’, that ‘In order to give an account of logic, we must presuppose and employ
logic.’26 Our semantic clause for object-language conjunction, ∧, always involves
conjunction in the metalanguage. On the Hybrid approach, our semantic clause
for object-language universal quantification, ∀, involved (metalinguistic) quantifi-
cation over all the ways in which a new name could be added to a signature. We do
not, of course, claim that anyone could read these semantic clauses and come to un-
derstand the very idea of conjunction or quantification from scratch. We are making
a much more mundane point: to understand the hybrid approach to semantics, one
need only understand a tractable number of ideas.
Second, in describing our semantics as compositional, we are not aiming to sup-
ply a semantics according to which the meaning of ∀xF(x) depends upon the sep-
arate meanings of the expressions ∀, x, F, and x.27 Not only would that involve an
oddly inflexible understanding of the word ‘compositional’; the discussion of §1.4
should have convinced us that variables do not have semantic values in isolation.28
Instead, on the hybrid approach, the meaning of ∀xF(x) depends upon the mean-
ings of the quantifier-expression ∀x…x and the predicate-expression F( ). The
crucial point is this: the Hybrid approach delivers the truth-conditions of infinitely
many sentences using only a small ‘starter pack’ of principles.
Having aired the virtues of the Hybrid approach, though, it is worth repeating
that our three semantic approaches are technically equivalent. As such, we can in
good faith use whichever approach we like, whilst claiming all of the pleasant philo-
sophical features of the Hybrid approach. Indeed, in the rest of this book, we simply
use whichever approach is easiest for the purpose at hand.
This concludes our discussion of first-order logic. It also concludes the ‘philo-
sophical’ component of this chapter. The remainder of this chapter sets down the
purely technical groundwork for several later philosophical discussions.
the two composite expressions ∀x and F(x). For if we think that open formulas possess semantic values
(in isolation), we will obtain an exactly parallel (and exactly as confused) ‘antinomy of the open formula’
as follows: clearly F(x) and F(y) are notational variants, and so should have the same semantic value; but
they cannot have the same value, since F(x) ∧ ¬F(y) is not a contradiction.
22 logics and languages
The trick is to add new semantic clauses for our second-order quantifiers. In fact,
we can adopt any of the Tarskian, Robinsonian, or Hybrid approaches here, and we
sketch all three (leaving the reader to fill in some obvious details).
Tarskian. Variable-assignments are the key to the Tarskian approach to first-order
logic. So the Tarskian approach to second-order logic must expand the notion of
a variable-assignment, to cover both relation-variables and function-variables. In
particular, we take it that σ is a function which assigns every variable to some entity
a ∈ M, every n-place relation-variable to some subset of M n , and every function-
variable to some function M n Ð→ M. We now add clauses:
M, σ ⊧ X n (t 1 , …, t n ) iff (t M,σ
1 , …, t n ) ∈ (X )
M,σ n M,σ
for any
L -terms t 1 , …, t n
M, σ ⊧ ∀X φ(X ) iff M, τ ⊧ φ(X n ) for every variable-assignment τ
n n
Hybrid. The key to the Hybrid approach to second-order logic is to define, up-
front, the three-place relation between M, a formula φ, and a relation (or function)
on M.29 We illustrate the idea for the case of relations (the case of functions is ex-
actly similar). Let S be a relation on M n . Let RS be an n-place relation symbol not
occurring in L . We define M[S] to be a structure whose signature is L together
M[S]
with the new relation symbol RS , such that M[S]’s L -reduct is M and RS = S.
Then where φ(X) is an L -formula with free relation-variable displayed, we define:
The three approaches ultimately define the same semantic relation. And we call the
ensuing semantics full second-order semantics.
The relative merits of these three approaches are much as before. So: the
Tarskian approach unhelpfully treats relation-variables as if they were varying pred-
icates; the Robinsonian approach forces us to stretch the idea of a language to break-
ing point; but the Hybrid approach avoids both problems and provides us with a
reasonable notion of compositionality. (It is worth noting, though, that all three
approaches effectively assume that we understand notions like ‘all subsets of M n ’.
We revisit this point in Part B.)
In essence, M rel
n serves as the domain of quantification for the n-place relation-
variables, and M fun
n serves as the domain of quantification for the n-place function-
variables. As before, though, we can make this idea precise using any of our three
approaches to formal semantics. We sketch all three.
Tarskian. Where M is a Henkin structure, we take our variable-assignments σ
to be restricted in the following way: σ assigns each variable to some entity a ∈ M,
each n-place relation-variable to some element of M rel
n , and each n-place function-
fun
variable to some element of M n . We then rewrite the clauses for the full semantics,
1.11. henkin semantics 25
M fun
n . We then offer these clauses:
We say that Henkin semantics is the semantics yielded by any of these three ap-
proaches, as applied to Henkin structures. Importantly, Henkin semantics gener-
alises the full semantics of §1.10. To show this, let M be an L -structure in the sense
of Definition 1.2. From this, define a Henkin structure N by setting, for each n < ω,
n = ℘(N ) and N n as the set of all functions N Ð→ N. Then full satisfaction,
N rel n fun n
When we want to draw the contrast, we call the (plain vanilla) Comprehension
Schema the Impredicative Comprehension Schema. But this will happen only
rarely; we only mention Predicative Comprehension in §§5.7, 10.2, 10.c, and 11.3.
We could provide a similar schema to govern functions. But it is usual to make
the stronger claim, that the following should hold in all structures (for every n):30
Choice Schema. ∀X n+1 (∀v∃y X n+1 (v, y) → ∃p n ∀v X n+1 (v, p n (v)))
To understand these axioms, let S be a two-place relation on the domain, and sup-
pose that the antecedent is satisfied, i.e. that for any x there is some y such that
S(x, y). The relevant Choice instance then states that there is then a one-place func-
tion, p, which ‘chooses’, for each x, a particular entity p(x) such that S(x, p(x)).
For obvious reasons, this p is known as a choice function. Hence, just like the Com-
prehension Schema, the Choice Schema guarantees that the domains of the higher-
order quantifiers are well populated.
This leads to a final definition: a faithful Henkin structure is a Henkin structure
within which both (impredicative) Comprehension and Choice hold.31
1.12 Consequence
We have defined satisfaction for first-order logic and for both the full- and Henkin-
semantics for second-order logic. However, any definition of satisfaction induces a
notion of consequence, via the following:
Definition 1.7: A theory is a set of sentences in the logic under consideration. Given a
structure M and a theory T, we say that M is a model of T, or more simply M ⊧ T, iff
M ⊧ φ for all sentences φ from T. We say that T has φ as a consequence, or that T
entails φ, or more simply just T ⊧ φ, iff: if M ⊧ T then M ⊧ φ for all structures M.
Note that this definition is relative to a semantics. So there are as many notions of
logical consequence as there are semantics.
Here are some examples to illustrate the notation. Consider the natural numbers
N and the integers Z in the signature consisting just of the symbol <, where this
is given its natural interpretation. It is easy to see that both structures satisfy the
following axioms:
∀x∀y∀z((x < y ∧ y < z) → x < z)
∀x(x ≮ x)
∀x∀y(x < y ∨ x = y ∨ y < x)
30 For more, see Shapiro (1991: 67).
31 See e.g. Shapiro (1991: 98–9).
1.13. definability 27
These are the axioms of a linear order. Let TLO be the theory consisting of just these
three axioms. Then we would write N ⊧ TLO and Z ⊧ TLO . But if we drop the third
axiom, we obtain the related notion of a partial order. For an example of a partial
order which is not a linear order, consider any set X with more than two elements,
and consider the structure P whose first-order domain is the powerset ℘(X) of X,
with < interpreted in P as the subset relation. If a, b are distinct elements of X, then
P ⊧ {a} ≮ {b} ∧ {a} ≠ {b} ∧ {b} ≮ {a}. So P ⊭ TLO .
1.13 Definability
In addition to a notion of consequence, a semantics will induce a notion of defin-
ability, as follows:
X = {(a1 , …, an ) ∈ M n : M ⊧ φ(a1 , …, an , b1 , …, bm )}
Here, the elements b1 , …, bm are called parameters. Many authors allow parame-
ters to be tacitly suppressed, and so say that X is definable iff X = {(a1 , …, an ) ∈
M n : M ⊧ φ(a1 , …, an )} for some φ(v1 , …, vn ) which is (tacitly) allowed to con-
tain further unmentioned parameters. If parameters are not allowed, such authors
typically say this explicitly. We will be similarly explicit. When parameters are not
allowed, the resulting sets are called parameter-free definable sets. Clearly a set is M-
definable iff it is parameter-free definable in some signature-expansion of M (see
Definition 1.4).
To illustrate the idea of definability, consider again the natural numbers N in the
signature consisting just of <, again with its natural interpretation. Here is a simple
definable set:
{0} = {n ∈ N : N ⊧ ¬∃x x < n}
Now, both of these sets are parameter-free definable. And so it follows that all defin-
able sets over N are parameter-free definable. For, where S is the successor function
28 logics and languages
on the natural numbers, each natural number n is equal to the term S n (0), which
we define recursively as follows:
(We label this definition ‘(numerals)’ for future reference.) Hence, to say that 2 =
S 2 (0) is just a fancy way of saying that two is the second successor of zero. The
terms S n (0) are sometimes called the numerals, and clearly N ⊧ n = S n (0) for
each natural number n ≥ 0. So, we can explicitly define the numerals in terms of
the less-than relation using G, any definable set on N is parameter-free definable, by
the following:
{(a1 , …, an ) ∈ N n : N ⊧ φ(a1 , …, an , b1 , …, bm )}
={(a1 , …, an ) ∈ N n : N ⊧ φ(a1 , …, an , S b1 (0), …, S bm (0))}
For an example of a structure with definable sets which are not parameter-free de-
finable, let L be a countable signature and let M be an uncountable L -structure.
Since there are only countably many L -formulas, there are only countably many
parameter-free definable sets. But trivially the singleton {a} of any element a from
M is definable, as {a} = {x ∈ M : M ⊧ x = a}. So M has uncountably many
definable subsets which are not parameter-free definable.
Finally, it is worth mentioning a particular aspect of definability in second-order
logic. Consider the natural numbers N in the full semantics, and consider the set
{(n, A) ∈ N × ℘(N) : N ⊧ A(n)} consisting of all pairs of numbers and sets
of numbers such that the number is in the set. It obviously makes good sense to
say that this set is definable, even though it is not a subset of N × N but rather of
N × ℘(N). So, in the case of second-order logic, we expand the notion of definabil-
ity to include both subsets of products of the second-order domain, and subsets of
products of the first-order domain and the second-order domain. This point holds
for both the Henkin and the full semantics.
Definition 1.9: The theory of Robinson Arithmetic, Q, is given by the universal closures
of the following eight axioms:
Another random document with
no related content on Scribd:
testes magyar történeti könyv nem lévén forgalomban. Amilyen
pompás részlet-munkálatok vannak a magyar történet területéről,
amely művek azonban, természetüknél fogva, nem jutnak a
nagyközönség elé, olyan árva az összefoglaló irodalom, amelyben
majdnem kizárólag csak iskolakönyvek, vagy iskolaszerü könyvek
jelennek meg. Egyik rosszabb és korlátoltabb a másiknál. Az
Akadémiának, noha ezt nem mondta, igaza van. Gyatrák e magyar
történetek és semmi egyébre nem alkalmasok, mint arra, hogy az
olvasót félrevezessék. Vezetik is. Nincs a magyar közép-
műveltségnek egyetlen olyan embere sem, aki tisztában volna akár
Nagy Lajos, akár Mátyás uralkodásával. Pedig e két alak nagyon
történeti s a szaktudomány előtt már a leghatározottabb
körvonalakkal áll. E szaktudományt azonban a történeti
kompendiumok irói nem hajlandók tudomásul venni. Példa: a
magyar honfoglalásnak van egy ragyogó történetirója, Sebestyén
Gyula a neve. Valósággal testi gyönyörüség elolvasni, hogy a
rendelkezésre álló gyér és keverék-adatokból ez az ember hogyan
hámozta ki egyrészt a magyar honfoglalásnak a történetét, másrészt
a monda-világát, hogyan választotta el a kettőt egymástól, hogyan
vitte tálcán be a történetbe azt, aminek csak kusza nyomai voltak
elszórva. Az iskolai történetirás egyszerüen nem vett tudomást
Sebestyén Gyula munkálkodásának eredményéről. És nem vett
tudomást Acsády munkálkodásáról, sőt, II. Józsefet illetően,
Marczaliéról sem. Sebestyén Gyula megmondja és bebizonyitja,
hogy ki volt Anonymus, csodaszépen mutatja ki, hogy a neve
Adorján volt, III. Béla kancellistája volt, püspök volt Erdélyben. Ön, jó
olvasóm, azonban még most is ugy tudja, hogy Anonymus
valamelyik Béla király névtelen jegyzője volt. Mert ön igy tanulta az
iskolában, s mert a fiától, aki most tanulja, ugyanigy értesül.
Az ön fia meg éppugy megtanulja a gyász-magyarokról, vagy
gaz-magyarokról szóló csacsiságot, ahogy ön is megtanulta. Lehelt
és Botondot is tanulja, valamint azt is, hogy az Árpád-királyok
énekmondói hegedősök voltak. Mindez nem igaz, nem igy igaz,
sokkal szebb és érdekesebb maga az igazság. De azt még nem
terjesztik. Az Akadémia sem, amely egyszerüen nem adja ki a
pályadijat, de nem néz körül s nem látja meg, hogy van már magyar
történetirás, amelynek kiadhatná.
A PERZSÁK
MÉLÁZÁS
GYORSÁRU
Updated editions will replace the previous one—the old editions will
be renamed.
1.D. The copyright laws of the place where you are located also
govern what you can do with this work. Copyright laws in most
countries are in a constant state of change. If you are outside the
United States, check the laws of your country in addition to the terms
of this agreement before downloading, copying, displaying,
performing, distributing or creating derivative works based on this
work or any other Project Gutenberg™ work. The Foundation makes
no representations concerning the copyright status of any work in
any country other than the United States.
• You pay a royalty fee of 20% of the gross profits you derive from
the use of Project Gutenberg™ works calculated using the
method you already use to calculate your applicable taxes. The
fee is owed to the owner of the Project Gutenberg™ trademark,
but he has agreed to donate royalties under this paragraph to
the Project Gutenberg Literary Archive Foundation. Royalty
payments must be paid within 60 days following each date on
which you prepare (or are legally required to prepare) your
periodic tax returns. Royalty payments should be clearly marked
as such and sent to the Project Gutenberg Literary Archive
Foundation at the address specified in Section 4, “Information
about donations to the Project Gutenberg Literary Archive
Foundation.”
• You comply with all other terms of this agreement for free
distribution of Project Gutenberg™ works.
1.F.
1.F.4. Except for the limited right of replacement or refund set forth in
paragraph 1.F.3, this work is provided to you ‘AS-IS’, WITH NO
OTHER WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED,
INCLUDING BUT NOT LIMITED TO WARRANTIES OF
MERCHANTABILITY OR FITNESS FOR ANY PURPOSE.